17
Communicating Uncertainty with Probability Phrases David V. Budescu Fordham University, NYC, USA IPCC Communication Meeting Oslo, Feb 9-10, 2016 I

Communicating Uncertainty with Probability Phrases

Embed Size (px)

Citation preview

Page 1: Communicating Uncertainty with Probability Phrases

Communicating Uncertainty with Probability Phrases

David V. Budescu Fordham University, NYC, USA

IPCC Communication MeetingOslo, Feb 9-10, 2016

I

Page 2: Communicating Uncertainty with Probability Phrases

Translation Table for IPCC Authors

Phrase Likelihood Conveyed

Virtually certain > 99%

Very likely > 90%

Likely > 66%

About as likely as not 33% to 66%

Unlikely < 33%

Very unlikely < 10%

Exceptionally unlikely < 1%5

Page 3: Communicating Uncertainty with Probability Phrases

Reservations

• The ranges of the terms are not mutually exclusive

• The ranges associated with the terms are arbitrary and are not related to people's “natural” understanding of the language

• People prefer to receive numerical information if available

• Large variability in interpretation of the terms

Page 4: Communicating Uncertainty with Probability Phrases

The 2014 Intl Study

• Read 8 statements extracted from the IPCC executive summary containing 4 terms: very likely, likely, unlikely, and very unlikely (2 statements for each term)

• Provide Best estimate and Lower and Upper bounds for each term (as intended by authors)

• Presentation format: Translation or Dual (VN)

• 27 samples in 25 countries and 17 languages (~400 /country) • All materials were translated and back translated by native

speakers• Analyze n=10,239 in 25 samples (51% men; mean age = 41)

7

Page 5: Communicating Uncertainty with Probability Phrases

Translation Screen shot

8

Page 6: Communicating Uncertainty with Probability Phrases

Translation screen shot with guidelines

9

Page 7: Communicating Uncertainty with Probability Phrases

Verbal-Numerical screen shot

10

Page 8: Communicating Uncertainty with Probability Phrases

Estimates are Regressive, but the VN Presentation Makes Them More Extreme

12

T VN T VN T VN T VNVery Unlikely

<10%Unlikely

<33%Likely>66%

Very Likely>90%

0

20

40

60

80

100

Bes

t Est

imat

es

Page 9: Communicating Uncertainty with Probability Phrases

Consistency with IPCC Guidelines Improves (especially for extremes)

13

Very Unlikely Unlikely Likely Very Likely0

10

20

30

40

50

16.13

36.62 37.64

18.95

31.76

43.1447.59

36.65

Overall TranslationOverall VN

% o

f Est

imat

es C

ompl

ying

with

G

uide

lines

Page 10: Communicating Uncertainty with Probability Phrases

IndiaNetherland

UKBrazil

Hong KongItaly

SpainCanadaQuebecSweden

AustraliaOverall

GermanyJapan

TaiwanRussiaPolandFrance

SlovakiaUSA

ChileChina

RSATurkeyKoreaIsrael

21%24%25%25%25%26%27%27%27%27%27%27%27%28%28%28%28%28%29%30%31%31%31%32%33%35%

28%32%

43%36%

41%39%

37%46%

38%41%

37%40%

43%35%

43%39%40%

38%44%45%

43%40%

48%46%

38%54%

% Compliance Translation Series2

14

Page 11: Communicating Uncertainty with Probability Phrases

The Benefits of Multiple Modalities

• The dual scale was superior in several senses: – Better differentiation and discrimination, – Higher agreement with the prescribed meaning, – More uniform interpretations across various groups– It works for people with different preferences

• It makes communicators more mindful of the terms’ intended meaning

• It can be used in more creative and flexible ways to accommodate, and signal, different levels of (im)precision in various cases

Page 12: Communicating Uncertainty with Probability Phrases

The Evidence Based Approach• Identify the target audience• Identify a subset of terms that are reasonable

candidates for inclusion in the lexicon• Establish empirically how the target population

understands, interprets and uses these terms (using single estimates, ranges, MFs)

• Use statistical methods to Identify optimal cutoff points that archive the highest consensus in interpretation in the target population

Page 13: Communicating Uncertainty with Probability Phrases

Evidence-Based Lexicons

Group Membership FunctionsDistribution of Estimates

Page 14: Communicating Uncertainty with Probability Phrases

Application to the IPCC

Sample Lexicon Consistency (%)UK IPCC 27

Peak 44Membership 50

Australia IPCC 25Peak 41Membership 45

Page 15: Communicating Uncertainty with Probability Phrases

Evidence-Based Lexicons

• There is a clear advantage to deriving evidence - based lexicons over “intuitive” ones

• Such lexicons are constructed to take full advantage of the common norms and meanings shared by a majority of participants in the target group

• As such they are less sensitive to idiosyncratic preferences of the committees that construct them and they are easier to implement and maintain

• There are many possible methods one can use in this context.

Page 16: Communicating Uncertainty with Probability Phrases

Evidence-Based Lexicons

• There is no single universal lexicon that serves all purposes

• It makes sense to choose the size and composition of the lexicon to – Fit the needs of the target population – Be sensitive to the precision and discrimination of

the information prevalent in the decision environment

– Be sensitive to the probabilities of interest

Page 17: Communicating Uncertainty with Probability Phrases

Thank you

[email protected]